Ledger — Finance Automation Workflow

Tags

The challenge

The finance team spent days each month manually categorizing transactions across multiple accounts, cross-checking them against expected patterns, and assembling a summary report by hand. Mistakes were rare but costly, and the manual process left little time for actual analysis.

The solution

• Designed a LangGraph-based agent pipeline that categorizes incoming transactions against the company's chart of accounts, with confidence scoring on every decision.
• Built anomaly detection that flags transactions deviating from historical patterns for human review, rather than silently auto-approving everything.
• Added a human-in-the-loop approval step in a purpose-built internal Next.js tool, so every automated categorization or flagged anomaly gets explicit sign-off before it touches the books.
• Automated the draft of the monthly financial summary from categorized data, leaving the finance lead to review and refine rather than assemble from scratch.

The results

• Manual reconciliation time dropped by roughly 70% in the first two months of use.
• Anomaly flags caught several transactions that would previously have gone unnoticed until a later audit.
• The finance lead now spends the freed-up time on forecasting and analysis instead of data entry.